Fine-grained image classification method based on generative adversarial network and attention network

Abstract

The invention discloses a fine-grained image classification method based on a generative adversarial network and an attention network, and the method comprises the steps: determining an image classification category, and building a training image set of a corresponding category; designing a deep attention convolution network for image fine-grained classification, wherein the network comprises fourparts, namely a VGG16 full convolution layer, SS attention region generation, a spatial pyramid pooling layer ROI pooling layer and an overall and local feature combined classification full connection layer; designing a structure of a generative adversarial network DAC-GAN, a generative network and a discrimination network; training a DAC-GAN network by using the training sample set, and storinga discriminant network model; and carrying out classification prediction on the image types by using the discrimination network model. According to the invention, the accuracy of the image classification network is improved, and the problem of insufficient data of a small sample size is solved.

Description

technical field [0001] The invention relates to a fine-grained image classification method, in particular to a fine-grained image classification method based on a generative confrontation network and an attention network, belonging to the field of image processing. Background technique [0002] With the deepening of deep learning technology research, convolutional neural network (CNN), as a kind of neural network, can extract and classify different features of the input, and the accuracy has been continuously improved, from 80% to 95%. The network structure of CNN consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. Among them, the convolutional layer extracts different features of the input through the convolution operation, the shallow convolutional layer extracts low-level features such as edges and lines, and the deep convolutional layer extracts advanced features; the pooling layer is connected after the convo...

AI Technical Summary

Problems solved by technology

However, this patent uses local image features and does not explain the principle of feature area acquisition. If there are too many feature areas, the overall calculation amount will be large; at the same time, too much emphasis is placed on local image features, which is not well compatible with the overall features.

Images